Researchers from the lab of Nobel laureate, David Baker, PhD, have now released RFdiffusion3 open source. The latest version of the de novo protein design model generates proteins that interact with any type of molecule, including DNA and small molecules with broad applications across sustainability and therapeutics.\u00a0<\/p>\n
\u201cWe built this as a general model and are\u00a0sharing\u00a0it freely so the scientific community can create things we haven\u2019t even imagined yet,\u201d said Rohith Krishna, PhD, postdoctoral researcher in the Baker lab and co-lead author of the\u00a0RFdiffusion3\u00a0bioRxiv\u00a0preprint<\/a>\u00a0that has not been peer reviewed.\u00a0<\/p>\n The work shows experimental proof of concept for the design of proteins that recognize specific sequences of DNA, which serves as a foundation for synthetic transcription factors for gene therapy, and complex enzyme catalysis, which offers industry applications.<\/p>\n Baker, who is the director of the Institute for Protein Design at the University of Washington (UW) and a Howard Hughes Medical Institute (HHMI) investigator, won the 2024 Nobel Prize in Chemistry for the design of novel proteins. RFdiffusion3 is one step closer to designing and understanding biological complexity, a challenge\u00a0Baker recently described to\u00a0GEN<\/a>\u00a0as one of the main obstacles to overcome for de novo design to transform medicine.\u00a0<\/p>\n RFdiffusion3\u00a0enters a growing ecosystem of protein design models\u00a0that cover multiple modalities, such as\u00a0BoltzGen<\/a>, which was released in October. BoltzGen demonstrated generalizable therapeutic design in any format, including nanobodies, mini-binders, and disulfide-bonded peptides, and against any target across nucleic acids, small molecules, and both ordered and disordered proteins.\u00a0Notably,\u00a0BoltzGen\u00a0designs\u00a0binders only and does not generate catalytic enzymes\u00a0as\u00a0demonstrated\u00a0by\u00a0RFdiffusion\u00a03.\u00a0<\/p>\n To achieve intricate chemical interactions with\u00a0new\u00a0precision,\u00a0RFdiffusion3\u00a0treats individual atoms as the fundamental units being designed,\u00a0a\u00a0methodological advance\u00a0from its predecessor,\u00a0RFdiffusion2<\/a>, concurrently published in\u00a0Nature Methods, which implements a hybrid atom and amino acid residue approach. In computational time, RFdiffusion3 exhibits ten-fold faster performance over RFdiffusion2, despite taking a more computationally intensive all-atom method.<\/p>\n Krishna told\u00a0GEN that the underlying intuition for RFdiffusion 3 was to give \u201cmore knobs for generality and controllability.\u201d As an example, the model\u2019s all-atom approach can allow the specification of a hydrogen bond to a particular atom, a level of precision that can greatly facilitate the construction complex reactions, such as enzyme catalysis. He also notes that RFdiffusion3 has limitations when designing against biomolecules not seen in the training set, such as non-canonical amino acids.\u00a0<\/p>\n Straight from the computer\u00a0<\/p>\n While RFdiffusion3\u2019s open-source launch opens the door for a new wave of enzyme innovation across the community, the Baker lab is already charting an impressive path forward in\u00a0complex\u00a0catalytic design.\u00a0<\/p>\n